feat: fuzzy search based on Levenshtein distance

[christoph-heinrich]

I don't know if that will ever go anywhere, but it's easier to test this way and it's a dedicated place to discuss this.

Ref. #625 (comment)

Depends on #644, but doesn't add text and hint characters together.

[tomasklaen]

I find it more confusing then helpful. For example, trying to match word upscale while leaving out letter p produces:

u - matches
us - nothing (I'd give up here and rewrite/fix my query)
usc - matches again

I'm really fine with what we have + #644. Implementing fulltext search features for a dumb single level list of short phrases is waay too much. I'd prefer a simple search implementation with slimmer menu.lua file over having this in. Please don't waste your time working on this :)

[christoph-heinrich]

It's less confusing without the cutoff.

[christoph-heinrich]

I like wasting my time 🙃

As it turns out Levenshtein is much slower then Hamming, and the query length makes a huge difference.
Reusing the matrix table gave a 4x performance improvement, and pulling out everything that can be prepared ahead of time gave another 2x improvement.
It should be fine for anything but extreme use-cases and I'd expect the query to typically be fairly short anyway.

That's probably as good as a fuzzy search gets.